Week2 Lecture3

W1L3
Clinical Epidemiology:
Defining Your Research
We can further classify the four different areas of clinical research that we defined in the
previous lecture into two distinct types of research. In general, clinical research questions can
be considered to be either descriptive or causal in nature. Simply put, in descriptive research,
the main aim is to describe a relationship between some determinants and a clinical outcome,
whereas in causal research the main aim is to find out whether the clinical outcome is directly
caused by the determinant or determinants you’re interested in. This distinction will be
important when considering what you plan to achieve with your research, as well as how you’ll
need to go about conducting your research.
So how does this relate to our DEPTh model? In general, for diagnostic and prognostic research
problems, we’re interested in seeing which information can help us predict a patient’s diagnosis
or prognosis. In effect, our aim is to describe the outcome in relation to some patient or test
characteristics, known technically as determinants or predictors. In diagnostic research, we’re
interested in predicting the presence of disease, and therefore we want our research to
describe how patient information and tests can help determine whether a patient currently has
a specific disease. In prognostic research we try to predict the future: we want our research to
describe how patient characteristics can help to predict the course of their disease.
In descriptive research any factor that’s related to the outcome may hold valuable predictive
information, and we might be interested in utilizing that information directly to enhance our
predictions. We’ll go into more detail about this in weeks two and three of the course.
When you’re interested in the etiology of a disease, you’re by definition interested in what the
cause of the disease is. The same goes for therapeutic, or intervention, research questions,
where you’re interested in whether the therapy of interest causes patients to get better, or
possibly even get worse. In this type of research, it’s not enough to just describe the
associations between an etiologic factor, or a treatment, and a health outcome; evidence that
the specific determinant of interest is directly causally related to the outcome needs to be
established. This implies that in such causal research it should be unequivocally shown that the
observed relationship between that determinant and the outcome is not explained by another
factor; i.e. by confounding.
Confounding bias is an essential consideration in causal research. Consider the following

example. We are interested in studying the effects of smoking on the risk of Parkinson’s
W1L3
disease. In order to establish whether or not smoking indeed causes Parkinson’s disease, we
need to make sure there are no differences between smokers and non-smokers regarding other
risk factors for Parkinson’s disease, such as age or alcohol consumption, or that these
differences are accounted for. Also in intervention research getting rid of such confounding is
crucial; this will be discussed during week four of this course. In descriptive, i.e. diagnostic and
prognostic, research causality and thus confounding is a non-issue: all determinants can be
used to predict the diagnostic or prognostic outcome.
Now that we have carefully categorized the nature of both our clinical problem and the
research we would like to conduct, we can further develop our research question into one that
is truly “water-tight”. We can do this by specifying our research in terms of a relationship
between the occurrence of an outcome and one or more determinants, within a clinically
relevant population, who we’ll refer to as the domain of our research. The definition of the
occurrence of an outcome as a function of one or more determinants, or the occurrence
relation, is referred to as the “theoretical design of a study”, and it’s the first step in designing a
clinical epidemiological study.
Let me clarify this using our COPD example. We have a tentative diagnostic research question
along the lines of “How can we use patient information and test results to determine whether
or not a COPD patient with shortness of breath also has heart failure?” To make our aims
clearer, we can further define our research by three terms.
First we have the outcome. In this case, heart failure, or the probability of the patient having
heart failure. In our research, we‘ll need to gather information about any confirmed diagnosis
of heart failure in the patients enrolled in our study.
Next we have the determinants. In diagnostic research there’s typically more than one
determinant. We want to know all of the relevant diagnostic tests that we could perform on our
COPD patient to help determine whether they have heart failure. This also includes the patient
history because knowing the patient’s age, gender and medical history is almost certainly going
to improve our ability to reach the correct diagnosis. Remember that because this research
question is a diagnostic question, we’re interested in a descriptive relationship between the
diagnostic determinants and the outcome. Thus causality does not play a role, and in turn we
don’t need to worry about confounding in our study.
And finally we have the research domain: i.e. the group of people to whom we may want to
apply our research findings. Careful consideration of the domain is vitally important as we need
to make sure that our study collects information that’s relevant to future patients. Do our
research findings need to be applicable to any future patient who enters the clinic? Not quite.
W1L3
Remember our initial patient, the 70 year old man who came into the clinic. He was suspected
of having heart failure but we needed clearer guidance to determine whether or not these
indications were instead signs of worsening COPD. Our clinical problem isn’t relevant to just
anyone, so instead we must focus on people with an established diagnosis of COPD that present
with symptoms suggestive of heart failure: this is the domain of our study.
By including these additional terms to our research question we complete the theoretical
design of our study. This enables us to truly cater the next two steps in designing a clinical
epidemiological study to our needs: the design of the collection of data and the design of the
analyses of the data.

Week2 Lecture3

Uploaded by

Copyright:

Available Formats

Week2 Lecture3

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Week2 Lecture3

Uploaded by

Copyright:

Available Formats

W1L3

Confounding bias is an essential consideration in causal research. Consider the following

You might also like